NEW FEATURES

• ggbarstats() (and grouped_ggbarstats()) now supports one-sample
  goodness-of-fit tests when only x is specified (with y = NULL as the new
  default). This produces a single stacked bar chart with chi-squared GOF test
  results in the subtitle, mirroring the existing one-sample support in
  ggpiestats() (#532, #697).

• ggpiestats() and ggbarstats() now compute pairwise contingency table
  analyses (Fisher's exact tests via pairwise_contingency_table()) when x
  has more than two levels. These are available via
  extract_stats(plot)$pairwise_comparisons_data. A new p.adjust.method
  parameter controls the p-value adjustment method (default: "holm").
  Pairwise results are not displayed on the plots since bar and pie charts lack
  a natural visual representation for pairwise significance annotations (#554).

• ggscatterstats() now sets a default number of bins for marginal
  histograms, suppressing the stat_bin() message about picking a better
  binwidth (#810).

• ggscatterstats() gains xsidehistogram.scale and ysidehistogram.scale
  parameters to control the scale (e.g., breaks, limits, transform) of
  the marginal distribution histograms (#898).

• All top-level plotting functions now include an alternative argument, which
  is passed down to {statsExpressions} to specify the alternative hypothesis for
  effect size confidence intervals (#794).

• ggbetweenstats() and ggwithinstats() now include a pairwise.alpha
  argument to control the alpha cutoff used for filtering displayed pairwise
  comparisons, and the secondary-axis label now reports the chosen alpha.

BREAKING CHANGES

• The following expert-level statistical parameters have been removed from all
  function signatures because their defaults are the universally recommended
  values and changing them requires specialist knowledge (#1087):

  • var.equal (from ggbetweenstats()): Welch's test (var.equal = FALSE)
    is uniformly recommended over Student's t-test.
  • nboot (from ggbetweenstats(), ggwithinstats()): 100 bootstrap
    resamples is adequate for trimmed-mean CIs.
  • sampling.plan, fixed.margin, prior.concentration (from
    ggpiestats(), ggbarstats()): Technical BayesFactor settings that
    virtually no one changes.
  • effsize.type (from ggbetweenstats(), ggwithinstats(),
    gghistostats(), ggdotplotstats()): The unbiased effect size estimator
    is now always used.

  Users who need non-default values for these parameters should call
  {statsExpressions} directly.

• ggwithinstats() (and grouped_ggwithinstats()) gains a subject.id
  parameter. When provided, the subject identifier column is used to correctly
  pair observations across conditions and to remove NA observations by subject
  key rather than by positional row number. Plots and statistical results for
  unsorted repeated-measures data will differ from previous versions once
  subject.id is supplied. All examples and vignettes have been updated to
  pass subject.id explicitly, which is now the recommended practice.

• The package argument has been removed from all plotting functions. The
  palette argument now accepts a single "package::palette" string (e.g.,
  palette = "ggthemes::gdoc"), matching the convention used by
  {paletteer} itself.

• The default palette has been changed from "RColorBrewer::Dark2" (8
  colors) to "ggthemes::gdoc" (24 qualitative colors), which accommodates
  categorical variables with up to 24 levels without errors (#1015).

• When the chosen palette does not have enough colors for the number of levels
  in the data, an error is now thrown (previously a warning was issued that
  was silently ignored until ggplot2 crashed anyway).

BUG FIXES

• Grouped plot functions (e.g., grouped_ggbarstats()) now preserve the order
  of groups as they appear in the data rather than sorting them alphabetically
  (#792).

• combine_plots() now renders the overall annotation title in bold by
  default, matching the styling used for individual plot titles.

• grouped_ggbarstats() and grouped_ggpiestats() now display a single
  unified legend when different groups have different observed factor levels for
  the x variable. Previously, patchwork could not merge the per-panel fill
  scales, producing duplicate legends (#868).

• ggbetweenstats() and ggwithinstats() now correctly display sample size
  labels on the x-axis even when centrality.plotting = FALSE (#695).

• ggcoefstats() now preserves the model term order in the default top-to-bottom
  plot layout and in estimate-sorted displays, instead of showing terms in the
  reverse order (#642).

• ggcoefstats() no longer draws empty stats.labels boxes for model terms
  whose label expression is absent, which affected mixed-model coefficient
  plots such as the documented lmer() example.